EP0232376B1 - Circulating context addressable memory - Google Patents
Circulating context addressable memory Download PDFInfo
- Publication number
- EP0232376B1 EP0232376B1 EP86905072A EP86905072A EP0232376B1 EP 0232376 B1 EP0232376 B1 EP 0232376B1 EP 86905072 A EP86905072 A EP 86905072A EP 86905072 A EP86905072 A EP 86905072A EP 0232376 B1 EP0232376 B1 EP 0232376B1
- Authority
- EP
- European Patent Office
- Prior art keywords
- sequence
- data
- symbols
- symbol
- memory
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/90335—Query processing
- G06F16/90344—Query processing by using string matching techniques
Abstract
Description
- The present invention is directed generally to the field of data processing systems and more particularly to memory systems for use in data processing systems.
- Essentially all data processing systems used in artificial intelligence systems require that information be stored and selectively retrieved. The information is normally organized as a database, stored in a memory, which consists of a number of records which are to be retrieved one at a time. Normally, each record is retrieved as a unit. The simplest example of a database retrieval involves retrieving all records that contain one or more key words. For example, the database might consist of all the articles in a particular set of scientific journals, one article per record. Here one would like to retrieve all articles containing one or more key words, such as all records containing the words "computing" and "artificial intelligence".
- In "expert systems" the database is codified into rules which must be searched repeatedly for rules conforming to a given specification. This given specification may be referred to as a "query sequence". This type of search, referred to as a "rule- based" search, differs significantly from the search for records containing one or more key words described above. When the expert system is given a problem, it must choose which of the many rules in the database to apply next to the problem at hand. This choice will depend on the data and relationships between the variables in the problem being solved. The expert system must select those rules which contain the same data and/or relationships between the variables. For example, these rules may be of the form:
If 〈condition〉 then 〈action or conclusion〉.
One of the rules in the data base might be as follows:
if (x eats) then (x is hungry). - If the problem contains the information "Peter eats", the system would be faced with the task of searching all the rules for those containing (x eats). Here x is a variable element which may be replaced by any element in the known information. The above rule together with any others involving "eats" would be retrieved. Once the system has retrieved the rule, it substitutes Peter for x and then concludes that Peter is hungry.
- The rule-based search differs significantly from the simple key word or key element search wherein all records containing one or more key words are retrieved. In a rule-based search, both the number and order of the elements contained in the database records are important. Consider a database containing the three records (a,x,b), (a,z,b), and (b,(d,e),a). A request for all records containing "a" and "b" would retrieve all of the above records. However, a request for all "rules" of the form (a,b) would not be satisfied by any of the records, since each record contains 3 elements and the requested rule contains only two. A request for all rules of the form (a,?,b), where ? denotes a variable element which may be satisfied by any constant, sub- expression, or another variable, would retrieve the first two records, but not the third, since "a" and "b" appear in the wrong order in the third record. Hence even with the concept of a variable, a rule-based retrieval system cannot conveniently be used to select records based on the key elements in an unspecified order.
- Conversely, a key word search system is difficult to use for rule-based searches. To perform a rule-based search on a key word system, one must use two steps. First, all records containing the elements in the rule are retrieved and stored in some auxiliary storage. Then each of these records must be separately examined to determine if it contains the elements in the proper order with no additional elements.
- A major problem encountered in implementing artificial intelligence systems using presently available computer hardware is the time required to search a large database for records matching a particular query sequence. To be useful, such a system must contain a very large number of records. Since the time needed to search the database is proportional to the length of the database, as a system having a fixed searching speed becomes "smarter" through the addition of new information in the form of new rules, it also become slower. This problem can only be solved by increasing the searching speed of the system as the size of the database to be searched is increased.
- In spite of the numerous improvements which have been made in integrated circuit technology, modern computers differ very little from the original Von Neumann designs. The classical Von Neumann machine consists of a central processing unit and a separate memory organized into words of fixed length. The processing unit fetches data from the memory, one word at a time by specifying the location of the desired word relative to the beginning of the memory. To perform the above search, a Von Neumann type central processing unit would sequentially retrieve each word of the database from the memory and would then compare it to each of the words in the query sequence which are also stored in the memory. Since there is a limit to the speed at which a single processing unit can run, Von Neumann type machines in practice are limited as to the size of the database that may be accessed. Even at speeds of 10 million compares per second, a Von Neumann machine would be hard pressed to search the contents of a library which contained , for example, only the law reporter series for the various jurisdictions in the United States.
- In addition to its speed limitations, the Von Neumann type architecture also presents a number of limitations because of its inherent hardware dependence. In order to specify a memory address, a fixed number of address lines are used. A system with N address lines can address a memory of 2N words and no more. To increase the memory capacity beyond this limit, one must increase the number of address lines. This involves both hardware and software alterations, since most systems specify a maximum memory size that the computer's instruction set is capable of addressing. Furthermore, if a portion of the memory is rendered inoperative by a malfunction of one of its components, it is difficult to transfer this storage to an undamaged segment of memory without reprogramming. As the size of the memory is increased in response to the need to store increased numbers of data records, the probability of such a malfunction in one component of the memory increases.
- Finally, a large database search problem will require the use of only a small fraction of the instructions present in the central processing unit. A typical central processing unit has literally hundreds of instructions that it is capable of executing varying from input/output instruction for dealing with the outside world to mathematics instructions for combining memory words which represent numbers. The database search problem requires at most 10 to 20 of these instructions. Hence the database search problem under utilizes the instruction repertoire of the central processing unit.
- The speed limitations of the central processing unit may be overcome to some degree by constructing a system having multiple central processing units wherein each has access to the memory. However, this solution has its limitations. The number of central procesing units that can share a given memory is ultimately limited by the time it takes each central processing unit to access the memory. If the memory bus must be dedicated to each central processing unit for 1/10th of the time, then no more than 10 central processing units can effectively share the same memory. Thus, replicating the central processing unit is not the best solution to the speed limitations of Von Neumann machines.
- Even if one could effectively reduce the search time using replicated central processing units, one is still left with the limitations imposed by the hardware. Sooner or later one will want to expand the database beyond these hardware limitations which is difficult to do in the Von Neumann type machines. Ideally, one would like to have a modular memory which could be added to a system whenever increases in the size of the database demand it without the concomitant necessity of increasing the number of address lines. This ability to expand the memory will become increasingly more important in those artificial intelligence systems where one seeks to build a machine which is capable of acquiring and using ever increasing amounts of information.
- Non-Von Neumann computing systems have been used in an attempt to improve the speed at which a computer can search a data base for a data sequence that matches a given query sequence. For example, The Colloquium on ASSOCIATIVE METHODS AND DATA BASE ENGINES, 17 May 1982,
pages 1/1-5/1, IEEE, London, GB, includes papers describing content addressable memories which apply parallel computing systems to the search problem. In one such sytem, the data sequences are stored in circulating memory array constructed from a long shift register. The query sequences are compared against the circulating data sequences at a number of different locations along the shift register. This type of system provides a potential improvement in the speed with which query sequences can be identified. Unfortunately, this type of prior art system is poorly suited to the types of comparisons required in rule-based systems. In particular, this type of system performs poorly in those situations in which a single symbol in one sequence is matched to a plurality of symbols in the other sequence. In addition, once a data sequence is identified, the time needed to output the sequence can be as long as the time needed for all the data sequences to pass some predefined point on the shift register. - The present invention provides a memory system for the storage and retrieval of data sequences of symbols, including a circulating memory comprising a first array of memory cells for storing one or more data sequences, each said data sequence consisting of one or more symbols, and means for causing said symbols to circulate sequentially past a plurality of tap points; means for receiving a query sequence of symbols coupled to said memory system; means comprising a second array of memory cells for storing said query sequence of symbols; and retrieving means operatively coupled to each said tap point for retrieving one or more sequences of symbols from said first array of memory cells which correspond to said query sequence of symbols, characterised in that said memory system further comprises means, coupled to said tap points, for said retrieving means to communicate with each other on an internal bus, wherein each time one of said retrieving means determines that a data sequence circulating past that retrieving means matches said query sequence, said retrieving means communicates the location of said data sequence on said internal bus to the other said retrieving means and the first said retrieving means to encounter that data sequence sends it over said internal bus for further processing.
- Thus, the invention consists of a memory system for storing and retrieving data sequences of symbols in response to a query sequence. Each of the data sequences and the query sequence is made up of three types of symbols, constants, delimiters, and variables. A data sequence is retrieved in response to a query sequence if the two sequences can be made identical by replacing each variable in each sequence by constants or combinations of constants and delimiters, said combinations beginning and ending with a delimiter. This determination is made by carrying out a number of pairwise comparisons between symbols chosen from each data sequence and the corresponding query sequence symbols. Those data sequences which successfully pass each of the pairwise tests are retrieved.
- The data sequences are stored in a circulating memory store in which each symbol is stored as a sequence of binary bits which periodically passes by a number of Tap points at which it is available for reading. Each of the Tap points contains a processor which is capable of comparing the stored data sequences with a query sequence consisting of a sequence of symbols of the same type as those used to represent the data stored in the circulating memory store. The memory system contains a separate storage buffer for holding this query sequence and translating it into a sequence of binary bits which may be conveniently compared to the sequence of binary bits representing the data sequences in the circulating memory store. A unique sub-system is provided for entering new data sequences into the circulating memory store without interrupting the operation of the data retrieval functions. This data entry technique automatically collects fragmented storage areas which were to small to store the new data sequence and combines these into a larger space into which the new data sequence is inserted. The operation of the various memory functions is under the control of a built-in processor which allows the system to be integrated into larger computer systems with a minimum of overhead.
- It is an object of the present invention to provide a memory system for the storage and retrieval of rules applicable to rule based artificial intelligence systems in which the rules are retrievable by specifying their content as opposed to their location in the memory.
- It is a further object of the present invention to reduce the time needed to retrieve said records by employing a replicated concurrent data processing system in which a plurality of processing units search for all data sequences satisfying a given query sequence.
- These and other objects will become clear from the following detailed description of the invention and from the accompanying drawings.
- FIGURE 1 is a block diagram of the preferred embodiment of the present invention.
- FIGURE 2 is a summary of the data formats used to store symbols in the preferred embodiment.
- FIGURE 3 is a block diagram of one of the Taps used in the preferred embodiment.
- FIGURE 4 is a block diagram of the query storage buffer used in the preferred embodiment.
- FIGURE 5 illustrates the insertion of new data into the memory in place of one or more blocks of unused space.
- FIGURE 6 is a state diagram for the finite state machine used to implement the Taps in the preferred embodiment.
- As pointed out above, the search time is proportional to the length of the database being searched. Hence the search time may be reduced by breaking the database into several subunits, each with its own processing unit. This is equivalent to dividing the search task between a number of small Von Neumann machines, each with sufficient memory to hold the key words in the search list and one or more words of the database. The fastest search with such a system would be realized with a memory which has one processing unit built into each word of memory. Then in one memory cycle, the memory could compare itself to a key word and identify each record in which the key word appeared. The optimum ratio of database words to processing units will clearly depend on the complexity of the processing unit relative to that of the memory. Since the processing unit needs only a few instructions to carry out the search function, a relatively high density of processing units is possible. This type of highly replicated modular structure is particularly well suited to modern VSLI integrated circuit fabrication techniques.
- This type of architecture is also well suited to the database search problem in that the processing time does not increase as the database size increases, since each unit of memory capacity carries its own searching capacity with it in the form of its built-in processing unit. Hence the computing capacity of the system expands as the database expands. Furthermore, the system's central processing unit, as opposed to that of the processing units which provide the search function in memory, has relatively little overhead in this type of design. It does not need to keep track of addresses or other data associated with the contents of the memory. It has only to send the query sequence to a memory unit made up of a number of modules with instructions as to where to store the results of the search.
- In principle, the data sequences can be stored in any of a number of physical memory devices. All of these devices include a memory of some sort for holding the data sequences and a processor for comparing these sequences with the query sequence. This comparison is carried out as a sequence of pairwise comparisons involving one symbol from the data sequence and one symbol from the query sequence. The time needed to search the stored data sequences for those that match a particular query sequence will depend on the speed of the comparator circuits which make these pairwise comparisons and the number of such circuits employed. Systems in which the ratio of comparators to memory locations is large will retrieve the matching sequences in the least amount of time; however, these systems will have a lower memory density and higher cost in general than systems with fewer comparators in which the circuitry that would have gone into additional comparators is converted to additional memory.
- In an embodiment of the present invention, a circulating memory storage system is used for the data sequences, the query sequence is held in a static memory, and the data sequences are matched against the query sequence as the data sequence circulates. Here, each data sequence is tested separately against the query sequence one at a time. If a particular data sequence matches the query sequence, it is readout before the processor goes on to test the next data sequence. If it does not match, the processor waits for the remainder of the sequence to pass, and then it tests the next stored data sequence.
- Most circulating memory systems are "bit serial" in nature. For example, magnetic disks generally read and write tracks which are a sequence of single bits obtained by converting the data words into a serial sequence of bits and then writing those bits into the circulating storage medium. There are two possible design strategies given the serial nature of the storage medium. First one could construct a system in which the serial bit stream is converted back into words, and the words then compared to the words in the query sequence. Such a system would use comparators which are one word wide and test one query word in a single "compare cycle". The second alternative would be to convert the query sequence words into a corresponding bit serial representation and then compare the two bit serial representations in a comparator which is only one bit wide. If the data words were N bits long, then N compare cycles would be needed; however, the cost of the comparators would be greatly reduced since the N bit wide comparator of the above scheme would be replaced by a single one bit wide comparator.
- Although the first of these alternatives would appear to provide faster comparisons, since it requires only one comparison cycle to test each query word against its corresponding data word as compared to the N comparison cycles needed by the second alternative, this is not in fact the case. The time required to make a comparison in the first case is the sum of the time needed to collect the N bits which make-up the data word from the bit serial stream and convert these into a data word which is then given to the N bit wide comparator. This operation would be carried out by shifting the data into a "serial-in parallel- out" shift register. Since the time needed to shift the N bits into a serial-in parallel-out shift register is at least as long as that needed to make N compares in a single exclusive OR gate, the first alternative takes at least as long to make a comparison as the simple one bit serial comparison algorithm. In addition, it is more costly, since it requires one serial-in parallel-out shift register per Tap point in addition to an N bit wide comparator. Hence, the bit serial approach is the method of choice for constructing a pattern addressable memory based on a circulating storage medium which is bit serial in nature.
- The memory system of the present invention is designed to retrieve all data sequences which correspond to a given query sequence. Each data sequence is made up of a sequence of symbols. Each symbol belongs to one of three possible types, constants, variables, and delimiters. The query sequence is likewise constructed from these three types of symbols. A stored sequence is defined to correspond to a query sequence if the two sequences can be made identical by replacing each of the variable elements in each of the sequences by a constant or some combination of constants which begins and ends with a delimiter. The constants or combinations thereof may be different for each of the variables replaced. For example, the sequence
(a,d,?,e,c,(a,b),g)
may be made identical to the sequence
(a,d,q,e,c,?,g)
by replacing the ? in the first sequence by q and the variable in the second sequence by the combination (a,b). - The matching algorithm is implemented as a series of pairwise matches between one element taken from the query sequence and one element taken from the data sequence being tested. At each successful pairwise match, the next elements to be used in the next match must be specified, since when sub-expressions are present, jumps around these expressions are often necessary. This may be illustrated using the above two sequences, the second sequence being the query sequence which is being matched against a data sequence represented by the first sequence. The match starts with "(" in the first sequence being matched against "(" in the second sequence. This process continues with the second and third elements. At the fourth match, the symbol "q" in the query sequence is matched against the variable "?" in the data sequence. Since "?" matches anything, this match is successful. The process continues until the "?" in the query sequence is matched against the "(" of the sub-expression contained in the data sequence. Since "?" matches the next element which in this case is the entire sub-expression, (a,b), the next element to be tested in the data sequence will be the "g" after the ")", not the "a" after the "(". Hence, when a variable in one sequence is matched against a "(" in the other sequence, the next symbols to test are the one following the "?" and the one following the ")" which corresponds to the "(" which was matched against the "?". A data sequence is said to correspond to the query sequence if each of the above pairwise matches is satisfied.
- In addition, the present invention implements an "almost exact" matching mode in which data sequences which begin with the sequence specified by the query sequence are returned even though they are longer than the query sequence. For example, if the query sequence (a,d) were used in the almost exact match mode, all data sequences which begin (a,d,.... would be retrieved. This mode allows one to retrieve data sequences of differing lengths which begin with the same sequence without specifying all possible combinations of ending sequences.
- The preferred embodiment of the present invention is shown schematically in FIGURE 1 at 10. It includes a circulating memory configurated as a bit
serial data stream 12 and a number of identicaldata processing units 18 which access this stream at a number of different Tap points, an example of which being shown at 22. - In the preferred embodiment, the bit
serial data stream 12 consists of a long shift register made-up of one bit memory cells. Each symbol of the stored data sequences is stored as binary coded number in a contiguous block of these one bit memory cells. The symbols stored in the shift register are made to circulate past each Tap point by shifting the contents of each memory cell to the next adjacent memory cell, the contents of the last memory cell being shifted into the first memory cell. - Other circulating memories which are more than one bit wide will be obvious to those skilled in the art. For example a shift register which is N bits wide where N is the number of bits used to store each symbol could be used to create a circulating memory in manner analogous to that described above using a one bit wide shift register.
- The preferred embodiment uses a fixed number of memory cells to store each symbol. Embodiments in which different numbers of memory cells are used to store different symbols will be obvious to those skilled in the art. In a system using variable numbers of memory cells per symbol, each symbol is preceded by a code indiciating the beginning of a symbol and a means for defining the end of the symbol is provided, e.g. a stop code.
- Each of the
processing units 18, hereinafter referred to as a Tap, is connected to aquery storage buffer 20 which contains the query sequence to be used in the search of the data sequences stored in said bitserial data stream 12. Aninput processor 16 is also attached to the bitserial data stream 12. This processor together with theinput buffer 14 is used to enter data into the bitserial data stream 12. The individual Taps communicate with each other on aninternal bus 26 which is also used to communicate with themaster control processor 24. Themaster control processor 24 communicates with the external processing system in which the memory of the present invention functions over anexternal bus 30. Themaster control processor 24 is also connected to thequery storage buffer 20 by abus 28 which is used to input the query sequence into thequery storage buffer 20. - Before a search can be initiated, the bit
serial data stream 12 must be loaded with the data sequences to be searched. This function is performed by theinput processor 16 which receives the data sequences to be stored from themaster control processor 24. The operation of this input system will be described in detail below. Individual data sequences can be added to the bitserial data stream 12 at any time. - Once the bit serial stream is loaded, a search for all data sequences corresponding to a given query sequence is carried out as follows. First, the query sequence is relayed to the
query storage buffer 20 by themaster control processor 24. Second, theTaps 18 and the stored data sequences circulating in the bitserial data stream 12 are initialized for a search as described below. Third, theTaps 18 are instructed to find a match to the query sequence in thequery storage buffer 20. Each time one of theTaps 18 finds a data sequence which matches the query sequence, it sends the location of that data sequence to theother Taps 18 on theinternal bus 26. Thefirst Tap 18 to encounter that data sequence then sends it overinternal bus 26 to the output handler which is contained in themaster control processor 24. Themaster control processor 24 relays the data sequence in question to the processing system in which the memory of this invention is operating. - Each data sequence is made up of a sequence of symbols. Special symbols are used to signal the beginning and end of each data sequence. Each symbol is represented by a storage "word" which is a contiguous block of one lost memory cells. In the preferred embodiment each storage word is 34 bits long. These 34 bits circulate in the bit
serial data stream 12 if the symbol is one of the symbols comprising a data sequence. Thequery storage buffer 20 contains 34 bit storage words for storing the query sequence symbols in the same 34 bit format. Each storage word is divided into an identification group and a data group. The first two bits of each word are identification bits which indicate the character of the data stored in the remaining 32 bits, referred to as the data bits. In storage words containing constants or variables, the data bits are used to store the "name" of the constant or variable. The name stored for a variable is not used by the apparatus of the present invention. This optional name is provided to make the data sequences read more like the English language. It may also be used by the data processing system coupled to the apparatus of the present invention. - In storage words which contain delimiters or which are used to signal some system function such as beginning and end of data sequence, one or more of the data bits may be used in conjunction with the identification bits to specify the function. The various identification bits and associated data bits are summarized in FIGURE 2. In addition to the special symbol for ending data sequences, there are special symbols, "seen" and "unseen", for indicating that a particular data sequence has been already tested in a match against the query sequence, and a special symbol, "empty" indicating that the storage word is available for use in storing a new data sequence. Either the seen or unseen symbol is used to indicate the beginning of a data sequence. When a search is initialized, the first symbol of each data sequence is changed to an "unseen" symbol. After a Tap compares this data sequence to the query sequence, the first symbol is changed to a "seen" symbol.
- The symbols which make-up the query sequence are coded in a similar format, as indicated in FIGURE 2. However, the concepts of seen and unseen data sequences and empty data storage words are not applicable here; hence only five types of data storage words are defined in the query sequence. These data storage words are the same as the corresponding storage words in the stored data sequences with the exception of the open delimiter.
- The data bits of a query sequence open delimiter are used to store the location of the corresponding closing delimiters. As pointed out above, in addition to a simple constant, a variable in a data sequence may match an entire sub-expression in the query sequence. Each such sub-expression begins with an open delimiter. Hence, this case is signaled by a variable in the data sequence being matched against an open delimiter in the query sequence. When an open delimiter in the query sequence is matched against a variable in the data sequence, the
Tap 18 making the match must jump to the closing delimiter in the query sequence corresponding to the open delimiter in question and then continue the matching operation starting with the query symbol after the closing delimiter. This is accomplished in the present invention by storing the location of the matching closing delimiter in the data bits of the query storage word containing the open delimiter. When an open delimiter in the query sequence matches a variable in the data sequence, theTap 18 jumps to the location in thequery storage buffer 20 specified in the data bits of the open delimiter and then continues the match using the symbol after the one specified in the jump. The jump location is encoded by setting one data bit in the open delimiter storage word for each symbol in the query sequence that is to be jumped over. Hence, if the query sequence contained the sub-expression (a,b), the open delimiter would have the first two data bits set to ones. - A block diagram for one of the
Taps 18 is shown in Figure 3. EachTap 18 includes a means for comparing a data sequence symbol with a query sequence symbol. The data sequence symbol is inputted to theTap 18 as a sequence of bits which are circulating in the bit serialserial data stream 12. The query sequence symbol to be used in the comparison is likewise inputted to theTap 18 in a bit serial format. The conversion of the query symbol to this format is performed in thequery storage buffer 20 as will be described below. The bit serial representation of the query symbol, referred to as the query stream, is inputted to the Tap online 56 which connects theTap 18 in question withquery storage buffer 20. - Each
Tap 18 contains 4 major elements. First, each Tap containsaddress logic circuitry 30 which keeps track of the "address" of the data sequence currently being read at saidtap point 22 associated with the Tap in question. This circuitry consists of anaddress counter 32, for computing the current address in the bitserial data stream 12 which is passing theTap 18, anaddress latch 34 for storing an address for communication to the other Taps on theaddress bus 36 which is part ofbus 26 which connects the various Taps in the system, and a comparecircuit 36 for testing the current address against one communicated on said address bus. Although the current address of a data sequence stored in the bitserial data stream 12 is not accessible outside of the apparatus of the present invention, internal to the apparatus, the data sequences are accessible by address. EachTap 18 computes its address relative to the other Taps and relative to some point in the bitserial data stream 12 which is redefined whenever theaddress circuitry 30 is initialized. This is accomplished by storing a predetermined number in each of the address counters 32 when theaddress circuitry 30 is initialized. This number is different for eachTap 18. It is the number of data bits in the bitserial data stream 12 between this Tap and the first Tap. As each data bit in the bitserial data stream 12 passes a given Tap, that Tap'saddress circuitry 30 increments theaddress counter 32 located in that Tap. If the operation being performed by aTap 18 requires it to communicate the address of a data sequence it has found to anotherTap 18, it does so by latching this address into theaddress latch 34 which is then gated onto theaddress bus 36 bygate 38. For example, when a match is found between a data sequence and the query sequence, theTap 18 that found the match relays this address of the data sequence in question to all of the other Taps. When anotherTap 18 encounters this address, it switches the bitserial data stream 12 to the output handler which is contained in themaster control processor 24. In the preferred embodiment, all of theTaps 18 share the same address bus, and a bus arbitration scheme is employed to resolve conflicts between two or more competing Taps. - The second major element present in each
Tap 18 is thedelimiter counter 42 used to find the closing delimiter corresponding to a given open delimiter marking a sub-expression in a data sequence. When a "?" in the query sequence is matched against a open delimiter in the stored data sequence, theTap 18 must wait until the corresponding closing delimiter is encountered in the data stream before resuming the comparison of the data and query sequences. This is accomplished by counting the delimiters that pass theTap 18 until the one corresponding to the open delimiter in question is encountered. Matching then resumes with the data sequence symbol following said closing delimiter. Since the sub-expression being skipped may contain another sub-expression as part of it, thedelimiter counter 42 must count both "(" and ")". Thedelimiter counter 42 is set to an initial state when a variable query symbol is compared to an open delimiter data sequence symbol. It is incremented on each "(" and decremented on each ")". So long asdelimiter counter 42 has a value different from the initial state, the circuitry which compares the query stream to the bit serial data stream is inhibited. When thedelimiter counter 42 returns to its initial state, the matching operation is resumed. The open delimiter marking the beginning of the sub-expression in question is also counted in this process. The delimiter counter acts both as a means for identifying the closing delimiter corresponding to given open delimiter in a data sequence and as a "hold" means for indicating that data symbol corresponding to a given query symbol is not yet available at theTap 18. - The third major element of each
Tap 18 is the comparing means for comparing the data and query sequence symbols. As will be discussed in detail below, thequery storage buffer 20 decodes the query sequence symbols into a bit serial stream having the same format as that used for the data sequence symbols circulating in the bitserial data stream 12. Since both streams are bit serial, a comparator in the form of an exclusive-OR gate 50 is sufficient to compare the two bit streams. The output of thegate 50 is connected to aflag 47 which is set to a first state when the first symbol of the data sequence being tested is compared to the first symbol of the query sequence. If any of the comparisons fails to find a match between the data sequence symbol being read at theTap 18 and the corresponding query symbol, theflag 47 is set to a second state. If the flag is still set to said first state after the last symbols of the data and query sequence are compared, the data sequence in question is to be retrieved. An indicator advancing circuit 44 monitors this flag, and matching is discontinued for the current data sequence whenever a mismatch is noted. - The fourth major element of each
Tap 18 is the indicator advancing circuit 44. The indicator advancing circuit 44 communicates with thequery store buffer 20 viasignal line 55 as discussed below and with themaster control processor 24 on theinternal bus 26. It also communicates with other Taps via theinternal bus 26 which contains theaddress bus 36 as a part thereof. - The main function of the indicator advancing circuit 44 is to supervise the comparison of the query sequence and data sequence serial bit streams and to specify the next query sequence symbol to be sent to the
Tap 18 from thequery store buffer 20. The symbols being compared in each bit stream can be divided into four types, constants, open delimiters, close delimiters, and variables, for the purpose of making this comparison. There are 16 possible combinations of pairwise comparisons to be made at theTap 18. In making these comparisons, the indicator advancing circuit first compares the identification bits of the symbols. These bits are stored in thequery status latch 46 and thedata status latch 48, respectively. If the identification bits indicate a possible match between the symbols, the indicator advancing circuit 44 goes on to compare the data bits of the symbols. - The possible outcomes of these comparisons are summarized in Table 1. In each case, the indicator advancing circuit 44 must determine if the match is to continue and, if it is to continue, which symbol of the query sequence is to be used at the next comparison. The choice of the next query sequence element is communicated to the
query store buffer 20 by signals on thecommunication line 55. Each time a pulse is sent online 55, a pointer, referred to as the query sequence pointer for this Tap, is advanced. Each Tap controls one query sequence pointer which specifies the query symbol to be sent to the Tap. If no pulses are sent, thequery storage buffer 20 repeats the bit serial stream for the current symbol. If the Tap sends one pulse on this line, thequery storage buffer 20 would send the next symbol in the query sequence. If two pulses were sent, thequery storage buffer 20 would skip one symbol in the query sequence and the next comparison would be carried out between the next symbol in the bitserial data stream 12 and the symbol stored two locations from the present one in thequery storage buffer 20. - In all but two cases, an examination of the identification bits and at most two of the following data bits in each stream will be sufficient for the indicator advancing circuit 44 to make its decision. These two cases involve the matching of a constant in the query stream against a constant in the bit
serial data stream 12 and the matching of an open delimiter in the query stream against a variable in the bitserial data stream 12. In the first of these cases (case 1 in Table 1), the indicator advancing circuit 44 must examine the entire data words in both streams to ascertain if the two constants match. If they do not, the matching process is discontinued for this data sequence. If they match, the query sequence pointer in thequery storage buffer 20 is advanced to indicate the next query sequence element by sending a pulse to thequery storage buffer 20 online 55. - In the second of these cases (case 15 in Table 1), a variable in the data sequence is matched against an open delimiter in the query sequence. Since this variable is to be matched with the entire subsequence which begins with this open delimiter, the query sequence pointer must be advanced to the corresponding close delimiter in the query sequence. As discussed in above, each open delimiter in the query sequence is coded with the number of symbols to be jumped to reach the corresponding close delimiter. This coding is in the form of a sequence of "1" bits in the data bits of the storage word used to store said open delimiter, one "1" for each symbol that must be skipped. Thus if the close delimiter were 4 symbols away in the query sequence, the open delimiter would contain 4 "1"'s. The indicator advancing circuit 44 merely relays these "1"'s to the
query store buffer 20 online 55. Circuitry in thequery storage buffer 20 advances the query sequence pointer. Once the query sequence pointer has been advanced to the corresponding close delimiter, the indicator advancing circuit 44 treats this case as if it involved a constants and a variable matched; i.e., the query sequence pointer is advanced one symbol in the query sequence and the matching continues with the next symbol in the bitserial data stream 12. - In the complementary case involving a variable in the query sequence which is matched against an open delimiter in the data stream (
case 12 in Table 1), the indicator advancing circuit 44 must wait until the corresponding close delimiter in the data stream passes before advancing the query sequence pointer. This is accomplished by counting the number of open and closed delimiters in thedelimiter counter 42 as discussed above. - In all of the 13 remaining cases, the match is either continued in which case the query sequence pointer is advanced to the next query sequence symbol, or the match is discontinued, in which case, the position of the query sequence pointer is left unchanged.
- In all cases (
Cases 5 through 8) in which a closing delimiter is encountered in the query sequence, the indicator advancing circuit 44 must check the data bits to see if this closing delimiter is in fact the end of query sequence marker. If it is said marker and the indicator advancing circuit 44 has been instructed to operate in the partial match mode described below, then the indicator advancing circuit 44 will treat this case as if the bitserial data stream 12 contained a closing delimiter indicating the end of the data sequence. If the indicator advancing circuit 44 is operating in the full match mode, a match will only occur if the data stream also contained an end of data sequence closing delimiter. The remaining cases listed in Table 1 are self explanatory. - In addition to supervising the matching functions, the indicator advancing circuit 44 also supervises the deleting and outputting of data sequences, initializing the data sequence identification bits, and initializing the memory as a whole. These operations are carried out in response to signals from the
master control processor 24. Deleting and outputting data sequences require that the indicator advancing circuit 44 match the address of the current data sequence being examined with that of the data sequence to be outputted or deleted. This address is placed on theaddress bus 36 as a result of a previous matching instruction. The instruction to delete or output is relayed by themaster control processor 24. When this address matches the address of the bit serial data sequence currently being read by saidTap 18 the appropriate action is taken. If the data sequence is to be outputted, it is copied to the output handler which is located in themaster control processor 24. If the data sequence is to be deleted, the identification bits and first two data bits of each word in the data sequence are set to the value indicating that the word is empty space. - To delete a specific sequence, that sequence is sent to the apparatus of the present invention as a query sequence. The matching operations results in this sequence being "retrieved," i.e. its address is placed on the
address bus 36. The delete instruction is then given. - Initializing the memory prior to a search for all data sequences that correspond to the query sequence involves setting the first two data bits of the first symbol of each data sequence. These bits are set to the "unseen" state discussed above. Initializing the memory as a whole is accomplished by setting the identification and first two data bits on every word to the "empty" state.
- A conventional random access memory has an instruction set consisting of only two commands, read and write. The apparatus of the present invention by virtue of its increased functionality has 7 commands which are summarized below. These operations are initiated by placing a code on
bus 30 which is used by themaster control processor 24 to communicate with the data processing system in which the present invention is functioning. Each code can be in the form of a special query symbol or signals on one or more control lines. The master control processor decodes these codes and initiates the proper actions internal to the apparatus of the present invention. - The FIND MATCH and FIND PARTIAL MATCH instructions are carried out in two steps. First, the query sequence is send to the apparatus of the present invention one symbol at a time on the
communication bus 30. Since the beginning and end of the query sequence are marked with special delimiter symbols, themaster control processor 24 can ascertain when the last symbol has been received. Once the last query symbol has been stored in thequery storage buffer 20, the appropriate bits of each storage word signaling the beginning of a data sequence in the memory are set to the "unseen" state, and the Taps are signaled to perform the matching operation. The FIND PARTIAL MATCH instruction differs from the FIND MATCH instruction only in that it returns all data sequences that begin with the query sequence as opposed to only those data sequences which match the query sequence. - Once a match is found, the GIVE MATCH instruction will result in the
master control processor 24 sending the data sequences found to the external data processing system onbus 30. If this operation is given when no match has been found, it is ignored. Each time the instruction is given to themaster control processor 24, one of the data sequences which matched the query sequence is transmitted to the external data processing system. - The DELETE MATCH instruction is used to delete a data sequence from the data store. It must be given after a GIVE MATCH instruction in which the data sequence which is to be deleted was sent as a query sequence. This delete instruction results in the identification and data bits of all the words in the data sequence whose address is currently latched on the
address bus 36 being set to the "empty" designation discussed above. - The ADD RECORD instruction is used to add a new data sequence to the bit
serial data stream 12. The actual insertion of the data sequence is carried out by the input buffer and itsrelated input processor 16 as described in detail below. Themaster control processor 24 merely relays the data sequence to be inserted to theinput buffer 14 on theinternal bus 26 as themaster control processor 24 receives it onbus 30. After the last symbol of the new data sequence is received, themaster control processor 24 signals theinput processor 16 to insert the new data sequence. - The INIT. MEMORY and CLEAR TAGS instructions result in the identification bits or the entire data words being cleared, respectively. These instructions are used to clear the memory before loading with the first data sequence.
- In the preferred embodiment of the present invention, the
query storage buffer 20 is a multi-ported storage buffer for storing the query sequence currently being used to select data sequences from the bitserial data stream 12. A single storage buffer is used to service all of theTaps 18 in order to minimize the storage space needed to hold the query sequence. In addition to storing the query sequence, the query storage buffer is responsible for providing a bit serial representation of the query symbols to thevarious Taps 18. EachTap 18 must have independent access to any of the symbols stored in thequery storage buffer 20. This is accomplished through an array of switches and pointers. EachTap 18 controls a pointer which specifies the word in the query storage buffer that is to be routed to it through the array of switches. - The
query storage buffer 20 used in the preferred embodiment of the present invention is shown schematically in Figure 4. It consists of amemory array 80 comprised of a plurality ofshift registers 84 as shown in FIGURE 4(a), each of which being used for storing one symbol from the query sequence currently being searched. The symbols are loaded into the memory array from the bus by themaster control processor 24. In the preferred embodiment, this is accomplished by shifting each successive symbol of the query sequence into the array of shift registers in the direction indicated by thevertical arrow 90. Each of the shift registers 84 recycles it contents by shifting in the direction Indicated by thehorizontal arrow 92 shown in FIGURE 4(a). This recycling operation creates a bit serial stream of data from each words This bit serial representation of each of the query symbols may be sent to one or more of theTaps 18. As bits are shifted off the end of each shift register they are used as input to the first bit in the register 88 and also as input to one or more of the Tap switches 82. There is one Tap switch 82 for eachTap 18. The choice of which word is send to aparticular Tap 18 is made by setting a query sequence pointer 94 associated with eachTap 18. - Each
Tap 18 is serviced by a tap switch 82 dedicated to thatTap 18 and which runs the entire length of thememory array 80. This switch allows thatTap 18 to receive any query sequence symbol as a bit serial stream. Theshift register 84 word that is switched to thatTap 18 is determined by a query sequence pointer 94 which is under the control of theTap 18 in question. Each query sequence pointer 94 consists of a shift register with one of the bits set to "1" and the remaining bits set to "0". The memory array location at which said "1" is present is connected to line 56 which is used to communicate a bit serial representation of the query sequence symbol located at that location to theTap 18 associated with said query sequence pointer 94. Each of these shift registers may be shifted by a signal from theTap 18 connected to said shift register by signals on aline 55 connecting saidTap 18 to thequery storage buffer 20. Each pulse sent results in the query sequence pointer 94 being shifted one word in thememory array 80. The query sequence pointers 94 are set to the first symbol in the query sequence as part of the initialization operations carried out at the beginning of each search under the control of themaster control processor 24 which communicates with the query storagebuffer control processor 96 overbus 28. EachTap 18 resets its associated query sequence pointer to the symbol marking the beginning of the query sequence prior to commencing a comparison with the first symbol of a data sequence being reat at thatTap 18. - Adding a data sequence to the store is complicated by the serial nature of the stored data sequences. Although the absolute location of an individual data sequence is not important, each data sequence must be stored in a contiguous block of storage locations. As data sequences are deleted and and new data sequences added to the bit
serial data stream 12, the free space may become fragmented and spread throughout the bitserial data stream 12 making it impossible to add a new data sequence even when sufficient free space exists. To correct this problem special hardware is provided to reclaim this space by consolidating free space so that it may be used to store new data sequences. This process is referred to as "Garbage Collection" below. - Garbage Collection is a problem associated with any storage system in which data is stored in data sequences which must occupy contiguous storage words. This problem is usually solved by recopying the data in the memory to a single contiguous block of storage words thus moving any free space between data sequences to one block of storage words, usually at the end of the space used by the data sequences already stored in the memory. This approach requires that the memory be taken out of service during the recopy operation. The present invention avoids this out of service time.
- In the present invention, the Garbage Collection function is combined with the data input function in a manner that allows Garbage Collection to take place automatically whenever new data sequences are stored in the bit
aerial data stream 12. The method uses a variable length loop of shift register storage, referred to as the "bubble", which may be inserted into the bitserial data stream 12. Referring to FIGURE 5(a), when thebubble 120 is inserted into the bitserial data stream 12 the circulating data in the bitserial data stream 12 leaves the bitserial data stream 12 at a specifiedpoint 100, enters thebubble 120 at apoint 102 which is determined by the contents of thebubble 120 as explained below, traverses thebubble 120, and re-enters the bitserial data stream 12 at thepoint 104 that it would have next traversed had thebubble 120 not been inserted. - A new data sequence is inserted in the bit
serial data stream 12 through thebubble 120. The data sequence is first loaded into thebubble 120 under the control of themaster control processor 24. Thebubble 120 is then inserted into the bitserial data stream 12 as shown in FIGURE 5(a). Referring to FIGURE 5(b), as the data in the bitserial data stream 12 circulates into thebubble 120 as indicated by the hatchedarea 106, the new data sequence stored in thebubble 120 is shifted into the bitserial data stream 12, as indicated by the cross- hatched area at 108. Data is shifted into and out of thebubble 120 as if thebubble 120 were a part of the bitserial data stream 12. - The length of the
bubble 120 may be changed by changing thepoint 102 in thebubble 120 at which the data leaving the bitserial data stream 12 enters thebubble 120. When empty storage words are shifted into thebubble 120, as shown at 110 in FIGURE 5(c), the entry point is moved from 102 to apoint 142 just ahead of the empty words. This leaves thebubble 120 in the configuration shown in FIGURE 5(d). This results in the bubble "shrinking" and the empty words being left in the unused portion of the bubble. When the bubble's length shrinks to zero, the input operation is completed. Hence, whenever a new data sequence is entered, and there is not sufficient contiguous space to store it, two or more smaller spaces will automatically be combined to create a space which is large enough for its storage, and this operation requires no more time than would be required to input the data sequence without garbage collection, since other memory functions can continue while the data in thebubble 120 is being inserted. - If a data sequence to be inserted is larger than the
bubble 120, it is broken into a series of short insertions, each of which being less than or equal to the maximum length of thebubble 120. After each insertion is made, theinput processor 14 which controls the bubble's operation records the location at which the last entry was made. Thebubble 120 is then reloaded and inserted when this location passes thebubble insertion point 104. - The apparatus of the present invention has been implemented on a single chip using standard CMOS integrated circuit techniques. The bit
serial data stream 12 memory is a 32,000 bit shift register constructed using CMOS integrated circuit techniques well known to those skilled in the art of integrated circuit fabrication. This specific embodiment contains 8Taps 18, and thequery storage buffer 20 has 36 words for storing the query sequence. - Each of the control circuits discussed above is constructed as a finite state machine in the form of programmed logic arrays using techniques well known to those skilled in the art of circuit fabrication. The indicator advancing circuit 44 in each of the
Taps 18 is typical. The algorithm executed by the indicator advancing circuit finite state machine is shown schematically in FIGURE 6. This machine can be separated into two parts. The first port is used to detect matches, and the second is used to report a successful match if one is found. Referring to FIGURE 6, the transition from the initial state, state 0, tostate 1 occurs when the matching function has been requested, the status identification bits of the present data word and query sequence word have been latched in thedata status latch 48 and thequery status latch 46, respectively, and these identification bits indicate that the data word is the start of an "unseen" data sequence. When these conditions are met, the state machine is initialized and advances to the matching state, markedstate 1 in FIGURE 6, and the address of the first symbol of the data sequence being examined is latched into theaddress latch 34 associated with thisTap 18. Matching then proceeds in one of two further states depending on the identification bits of the data and query words in question. If either the data or query word is a closing delimiter, then the machine advances tostate 2 and tests are made for the end of the matching operation or other condition associated with a match to a close delimiter. If neither word involves a closing delimiter, the state machine advances tostate 3 where the two words are matched against each other. If the results of the operations instate 3 indicate that the words match, the next query sequence element to be used in the match is specified, and the machine returns tostate 1 to await the next match operation between a query sequence word and the next data word in the circulating memory. If the data word is found not to match the query word, the machine returns to state 0. If a match is found, the state machine advances to the second section where matches are reported. - One problem inherent in a circulating storage system is that by the time the data sequence has been determined to match the query sequence the data sequence, has passed the
Tap 18 at which the comparisons were made, and hence is not available for read-out at thisTap 18 until it recirculates. The solution to this problem employed in the present invention is to store the starting address of the data sequence being checked when the machine advances from state 0 tostate 1 as part of the initialization function. If the data sequence is found to be a match, then this address is communicated to theother Taps 18. When anotherTap 18 reads the data sequence at this address, it copies the data stream to the output handler in themaster control processor 24. Hence the data sequences may be read out before it recirculates past theTap 18 at which the match was made. Since allTaps 18 share acommon address bus 36 for purposes of this communication, an arbitration scheme similar to that used in the UNIBUS arbitration scheme is used to prevent conflicts on theaddress bus 36. This scheme is shown instates Tap 18 is given a "grant" which it acknowledges by the "Sack" signal. As soon as the bus is free, theTap 18 takes control of theaddress bus 36 by generating a bus busy signal. Once theTap 18 has taken control of theaddress bus 36, the address of a matching data sequence will be held on theaddress bus 36 until the data sequence has been found and output is complete. - Methods for translating the finite state machine logic described above and in FIGURE 6 into a programmed logic array for executing the algorithm described above are well known to those skilled in the art of VLSI integrated circuit technology (c.f. J. Millman, MICROELECTRONICS, McGraw-Hill Book Co., ). Once the finite state machine functions are defined, a program for a programmed logic array is written. This program may then be implemented in any of a number of systems used to construct such arrays in VLSI circuitry.
- A programmed logic array program for implementing the indicator advancing circuit 44 functions discussed above and shown schematically in FIGURE 6 is shown in Table 2 as an example of how the functions performed by the various finite state machines used in the preferred embodiment may be implemented. The inputs to the indicator advancing circuit programmed logic array are shown along the top of Table 2 separated by a vertical line from the outputs. Those inputs and outputs which only function in
states query status latch 46 shown in FIGURE 3. Pbit and Dbit refer to the current bits in the two bit serial streams which are being compared. The bits associated with the data sequence symbol being compared are denoted by "D", D0 and D1 being the identification bits of the current data symbol which have been latched in thedata status latch 48 shown in FIGURE 3. The flag is theflag 42 which monitors the output of the exclusive-OR gate 50 shown in FIGURE 3. This flag stores the result of the last comparison mode by the exclusive-OR gate when the "WE" output signal was "1". The SOR input line indicates that a start of data sequence symbol has been seen by theTap 18 while themaster control processor 24 has requested that a match be made. The "STATUS" input line indicates that the identification bits D0, D1, P0, and P1 have been latched. The "SLOT" input indicates that a new word in the data stream has begun. Finally, the "J=0" line indicates that a sub-expression in the data stream is being skipped by counting open and close delimiters in thedelimiter counter 42 shown in FIGURE 3. This input is set to 1 by the initialization signal, INIT., and set to 0 by COUNT output. The INCR. and DECR. outputs result in this counter being incremented and decremented, respectively. When the counter returns to its initial value, "J=0" is reset from 0 to 1. - The inputs and outputs used exclusively in
states Tap 18. If the bus has not been granted, the program maintains the "FAM" signal requesting a bus grant. Once the bus has been granted, theTap 18 acknowledges the grant by the "Sack" output and entersstate 5 where it waits for the bus to be free as indicated by the "Bbsy" input. Once the bus is free, theTap 18 takes control of the bus through the "Bbsy" output which remains on until themaster control processor 24 signals that the data sequence has in question has been outputted through the "Ssync" input. - For simplicity, the incomplete match mode in which data sequences which are longer than the query sequence but which begin with the query sequence are also returned has not been shown in the programmed logic array program. Similarly, the programming for the various non-matching functions carried out by the Tap (delete data sequence, initialize memory, etc) have been omitted.
- The logic circuits used in the
input processor 16,master control processor 24, andquery storage buffer 20 are similarly constructed as finite state machines. Detailed state diagrams for these machines will be obvious to those skilled in the art of VSLI integrated circuit technology. - While various embodiments of the invention have been discussed herein, it will be appreciated that various changes and modifications may be made without departing from the present invention as claimed.
Claims (18)
- A memory system (10), for the storage and retrieval of data sequences of symbols, including a circulating memory (12) comprising a first array of memory cells for storing one or more data sequences, each said data sequence consisting of one or more symbols, and means for causing said symbols to circulate sequentially past a plurality of tap points (22); means (24) for receiving a query sequence of symbols coupled to said memory system (10); means (20) comprising a second array of memory cells for storing said query sequence of symbols; and retrieving means (18) operatively coupled to each said tap point for retrieving one or more sequences of symbols from said first array of memory cells which correspond to said query sequence of symbols, characterised in that said memory system (10) further comprises means (30), coupled to said tap points, for said retrieving means to communicate with each other on an internal bus (26, 36), wherein each time one of said retrieving means (18) determines that a data sequence circulating past that retrieving means (18) matches said query sequence, said retrieving means (18) communicates the location of said data sequence on said internal bus (26, 36) to the other said retrieving means (18) and the first said retrieving means (18) to encounter that data sequence sends it over said internal bus (26, 36) for further processing.
- A memory system as claimed in Claim 1, comprising means (14) for receiving a data sequence of symbols coupled to said circulating memory (12) and for storing said received data sequence of symbols in said first array of memory cells.
- A memory system as claimed in Claim 1 or Claim 2, wherein said retrieving means (18) retrieves all data sequences of symbols which are identical to said query sequence of symbols.
- A memory system as claimed in Claim 1 or Claim 2, wherein said data sequences of symbols and said query sequence of symbols are comprised of symbols of three types, namely delimiters, constants and variables and wherein said retrieving means (18) retrieves all data sequences of symbols which correspond to said query sequence of symbols, a data sequence of symbols being defined as corresponding to said query sequence if the two sequences can be made identical by replacing each variable in each of said sequences by a constant or combination of constants and delimiters, said combination beginning and ending with a delimiter.
- A memory system as claimed in Claim 4, wherein said retrieving means (18) comprises a plurality of tap point processors respectively operatively connected to said tap points (22), and each said processor comprising first indicator means (32) for identifying a data sequence symbol; second indicator means (94) for identifying a query sequence symbol; comparing means (50) for comparing the data sequence symbol identified by said first indicating means (32) with the query sequence symbol identified by said second indicator means (94) and for generating an output signal having one of two alternative states, a matched state indicating that the two symbols compared are the same or that at least one of said symbols is a variable, and a not-matched state indicating that the two symbols are different and neither is a variable; flag means (47) responsive to said comparing means output signal for indicating the generation of any output signal being in the not-matched state; indicator advancing means (44) comprising means for causing said first indicator means to indicate the first symbol of a selected data sequence, said second indicator means (94) to indicate the first symbol of said query sequence, and said comparing means to compare said identified symbols; and means for causing said first indicator means (32) to identify a next data sequence symbol, said second indicator means (94) to identify a next query sequence symbol, and said comparing means (50) to compare said identified symbols until a termination condition is detected; means for detecting a termination condition; and means (30) responsive to said termination condition for outputting said selected data sequence if said flag means (47) indicates that none of said comparisons resulted in the generation of a not-matched output signal.
- A memory system as claimed in Claim 5, wherein said termination condition is defined to occur when said comparing means (50) compares the last symbol of said query sequence with a data sequence symbol.
- A memory system as claimed in Claim 5, wherein said termination condition is defined to occur when said comparing means (50) compares the last symbol of said selected data sequence with a query sequence symbol.
- A memory system as claimed in Claim 5, wherein said termination condition is defined to occur when said comparing means (50) compares the last symbol of said selected data sequence with the last symbol of said query sequence, or when an output signal in the not-matched state is generated.
- A memory system as claimed in any one of Claims 5 to 8, wherein said delimiters occur in corresponding pairs, an open delimiter being used to mark the beginning of a sequence of symbols or of a sub-sequence of symbols embedded in a sequence of symbols, and a corresponding closing delimiter being used to mark the end of said sequence of symbols or sub-sequence of symbols, wherein said indicator advancing means includes means (42) for determining which closing delimiter corresponds to each open delimiter, and wherein said next data sequence symbol and said next query sequence symbol caused to be identified by said indicator advancing means (44) are the next adjacent data sequence symbol and query sequence symbol, respectively, unless the previous comparison was between a variable and an open delimiter in which case said next symbols are the next adjacent symbol after said variable and the next adjacent symbol after the closing delimiter corresponding to said open delimiter.
- A memory system as claimed in Claim 9, wherein said means for determining which closing delimiter corresponds to a given open delimiter comprises means for storing as part of each open delimiter the location of its corresponding closing delimiter in said sequence of symbols.
- A memory system as claimed in Claim 10, wherein each symbol is stored in a contiguous block of one bit memory cells, said block being divided into an identification group and a data group of cells, said identification group consisting of two or more cells specifying the type of data symbol stored in said block of one bit memory cells and said data group comprising the remainder of said block of one bit memory cells, and wherein the means of storing the location of a closing delimiter corresponding to any given open delimiter comprises setting one bit in the data group of the block used to store an open delimiter to a "1" for each symbol which must be jumped to reach the closing delimiter corresponding to said open delimiter in the sequence of symbols containing said open delimiter.
- A memory system as claimed in Claim 9, wherein said means for determining which closing delimiter in a data sequence of symbols corresponds to a given open delimiter in a said sequence of symbols, comprises means (42) operatively connected to each said tap point processor (18) for counting delimiters in the data sequence of symbols containing said open delimiter, said counting means being initialized to a predetermined first state when said open delimiter is read by said reading means, said counting means (42) being incremented each time an open delimiter is read in said data sequence of symbols containing said open delimiter, and said counting means (42) being decremented by an equal amount each time a closing delimiter is encountered in said sequence of symbols, said corresponding closing delimiter being the one causing said counting means (42) to indicate a second predetermined state.
- A memory system as claimed in any one of Claims 9 to 12, wherein said first indicator means (44) comprises a hold means for inhibiting said comparing means (50), said hold means having two states, namely ready or not ready, said not ready state indicating that the data sequence symbol to be compared to the query sequence symbol specified by said second indicating means is not the next symbol to be read at said tap point (22) and that said comparing means (50) is to be inhibited, and said ready state indicating that the data sequence symbol to be compared to the query sequence symbol specified by said second indicating means (94) is the next data sequence symbol to be read at said tap point, said indicator advancing means (44) setting the state of said hold means to the not ready state when a variable query sequence symbol is compared by said comparing means (50) to an open delimiter data sequence symbol, and said indicator advancing means (44) resetting said hold means to the ready state when the corresponding closing delimiter is read at said tap point (22).
- A memory system as claimed in Claim 11, wherein said second indicator means comprises a pointer (94) in said query sequence storing means (20), and said indicator advancing means (44) includes means for incrementing said pointer, said pointer being incremented after each comparison between a data sequence symbol and a query sequence symbol which resulted in said matched output being generated by said comparison means (50), and when an open delimiter query sequence symbol is compared with a variable data sequence symbol, said pointer (94) being incremented once for each data group bit set to a "1" in the data group of the block of one bit storage cells used to store said open delimiter.
- A memory system as claimed in Claim 2, or any one of Claims 3 to 14 as dependent upon Claim 2, wherein said received data sequence storage means (12) comprises a memory segment including a third array of memory cells (120) and means for inserting said memory segment into said circulating memory (12) at a predetermined exit point (100) such that data symbols leave said circulating memory (12) at said exit point (100), traverse said memory segment, and then re-enter said circulating memory (12) at the memory cell (104) immediately after said exit point (100) that said symbols would have traversed in said circulating memory (12) had said memory segment not been inserted; and wherein the number of memory cells in said memory segment may be varied by varying the location of the memory cell in said segment (120) at which data enters said memory segment in response to the symbols stored in said segment, said entry location being changed each time a symbol which is to be eliminated from said circulating memory (12) enters said memory segment at a point ahead of said symbol.
- A memory system as claimed in Claim 15, comprising means for inserting said memory segment into said circulating memory when a predetermined sequence of data symbols passes said exit point.
- A memory system as claimed in Claim 15 or Claim 16, comprising means for inserting a sequence of symbols into said memory segment in response to a signal coupled to said memory system.
- A memory system as claimed in any one of Claims 15 to 17, wherein said first and third memory arrays comprise one bit wide shift registers.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
AT86905072T ATE91815T1 (en) | 1985-08-13 | 1986-08-13 | CONTEXTUAL ADDRESSABLE Circulating Memory. |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US76539185A | 1985-08-13 | 1985-08-13 | |
US765391 | 1985-08-13 |
Publications (3)
Publication Number | Publication Date |
---|---|
EP0232376A1 EP0232376A1 (en) | 1987-08-19 |
EP0232376A4 EP0232376A4 (en) | 1989-05-16 |
EP0232376B1 true EP0232376B1 (en) | 1993-07-21 |
Family
ID=25073443
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP86905072A Expired - Lifetime EP0232376B1 (en) | 1985-08-13 | 1986-08-13 | Circulating context addressable memory |
Country Status (7)
Country | Link |
---|---|
US (1) | US4924435A (en) |
EP (1) | EP0232376B1 (en) |
JP (1) | JP2516609B2 (en) |
AT (1) | ATE91815T1 (en) |
CA (1) | CA1266330A (en) |
DE (1) | DE3688737T2 (en) |
WO (1) | WO1987001222A1 (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5604902A (en) * | 1995-02-16 | 1997-02-18 | Hewlett-Packard Company | Hole plugging garbage collection for a data storage system |
US6442543B1 (en) * | 1997-07-25 | 2002-08-27 | Amazon.Com, Inc. | Method and apparatus for changing temporal database information |
US5999924A (en) * | 1997-07-25 | 1999-12-07 | Amazon.Com, Inc. | Method and apparatus for producing sequenced queries |
US6185556B1 (en) * | 1999-05-04 | 2001-02-06 | Amazon.Com, Inc. | Method and apparatus for changing temporal database |
US7711763B2 (en) * | 2001-02-21 | 2010-05-04 | Mips Technologies, Inc. | Microprocessor instructions for performing polynomial arithmetic operations |
US7181484B2 (en) * | 2001-02-21 | 2007-02-20 | Mips Technologies, Inc. | Extended-precision accumulation of multiplier output |
US7162621B2 (en) | 2001-02-21 | 2007-01-09 | Mips Technologies, Inc. | Virtual instruction expansion based on template and parameter selector information specifying sign-extension or concentration |
US7599981B2 (en) | 2001-02-21 | 2009-10-06 | Mips Technologies, Inc. | Binary polynomial multiplier |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3245052A (en) * | 1962-05-17 | 1966-04-05 | Rca Corp | Content addressed memory |
US3701980A (en) * | 1970-08-03 | 1972-10-31 | Gen Electric | High density four-transistor mos content addressed memory |
US3906455A (en) * | 1974-03-15 | 1975-09-16 | Boeing Computer Services Inc | Associative memory device |
US4037205A (en) * | 1975-05-19 | 1977-07-19 | Sperry Rand Corporation | Digital memory with data manipulation capabilities |
US4027288A (en) * | 1976-02-09 | 1977-05-31 | Burroughs Corporation | Self-managing variable field storage system for handling nested data structures |
US4118788A (en) * | 1977-03-07 | 1978-10-03 | Bell Telephone Laboratories, Incorporated | Associative information retrieval |
US4283771A (en) * | 1978-07-31 | 1981-08-11 | International Business Machines Corporation | On-chip bubble domain relational data base system |
US4451901A (en) * | 1982-01-21 | 1984-05-29 | General Electric Company | High speed search system |
US4527253A (en) * | 1982-05-28 | 1985-07-02 | Hitachi, Ltd. | Data searching apparatus |
US4554631A (en) * | 1983-07-13 | 1985-11-19 | At&T Bell Laboratories | Keyword search automatic limiting method |
-
1986
- 1986-08-12 CA CA000515787A patent/CA1266330A/en not_active Expired - Fee Related
- 1986-08-13 EP EP86905072A patent/EP0232376B1/en not_active Expired - Lifetime
- 1986-08-13 AT AT86905072T patent/ATE91815T1/en not_active IP Right Cessation
- 1986-08-13 DE DE86905072T patent/DE3688737T2/en not_active Expired - Fee Related
- 1986-08-13 JP JP61504440A patent/JP2516609B2/en not_active Expired - Lifetime
- 1986-08-13 WO PCT/US1986/001683 patent/WO1987001222A1/en active IP Right Grant
-
1988
- 1988-05-02 US US07/188,640 patent/US4924435A/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
JP2516609B2 (en) | 1996-07-24 |
ATE91815T1 (en) | 1993-08-15 |
US4924435A (en) | 1990-05-08 |
EP0232376A4 (en) | 1989-05-16 |
WO1987001222A1 (en) | 1987-02-26 |
DE3688737D1 (en) | 1993-08-26 |
DE3688737T2 (en) | 1994-03-10 |
CA1266330A (en) | 1990-02-27 |
EP0232376A1 (en) | 1987-08-19 |
JPS63500547A (en) | 1988-02-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP0233272B1 (en) | Rule-based data retrieval method and apparatus | |
US4053871A (en) | Method and system for the iterative and simultaneous comparison of data with a group of reference data items | |
Kohonen et al. | Logic Principles of Content-Addressable Memories | |
US4747072A (en) | Pattern addressable memory | |
US4870568A (en) | Method for searching a database system including parallel processors | |
KR940003700B1 (en) | Method and apparatus for search | |
US5758148A (en) | System and method for searching a data base using a content-searchable memory | |
US6760821B2 (en) | Memory engine for the inspection and manipulation of data | |
JPS5846742B2 (en) | Interactive data search device | |
JPH0285927A (en) | Memory device | |
US4805093A (en) | Content addressable memory | |
US4283771A (en) | On-chip bubble domain relational data base system | |
GB1470104A (en) | Stored programme electronic computer | |
EP0232376B1 (en) | Circulating context addressable memory | |
Robinson | Pattern-addressable memory | |
US3548385A (en) | Adaptive information retrieval system | |
Healy | A character-oriented context-addressed segment-sequential storage | |
Robinson | The pattern addressable memory: Hardware for associative processing | |
Natarajan et al. | A multiaccess associative memory | |
US3344402A (en) | Multiple section search operation | |
Dattagupta | An investigation to study the feasibility of on-line bibliographic information retrieval system using an APP | |
Su | Cdllular § Logic'Devic Corj, ceptsaijdApplication | |
JPH043251A (en) | Method and processor for retrieving document | |
GB1373985A (en) | Index system | |
Lochovsky | DL Lee |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
17P | Request for examination filed |
Effective date: 19870406 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AT BE CH DE FR GB IT LI LU NL SE |
|
A4 | Supplementary search report drawn up and despatched |
Effective date: 19890516 |
|
17Q | First examination report despatched |
Effective date: 19910717 |
|
GRAA | (expected) grant |
Free format text: ORIGINAL CODE: 0009210 |
|
AK | Designated contracting states |
Kind code of ref document: B1 Designated state(s): AT BE CH DE FR GB IT LI LU NL SE |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: IT Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRE;WARNING: LAPSES OF ITALIAN PATENTS WITH EFFECTIVE DATE BEFORE 2007 MAY HAVE OCCURRED AT ANY TIME BEFORE 2007. THE CORRECT EFFECTIVE DATE MAY BE DIFFERENT FROM THE ONE RECORDED.SCRIBED TIME-LIMIT Effective date: 19930721 Ref country code: AT Effective date: 19930721 Ref country code: BE Effective date: 19930721 Ref country code: NL Effective date: 19930721 Ref country code: CH Effective date: 19930721 Ref country code: LI Effective date: 19930721 Ref country code: SE Effective date: 19930721 |
|
REF | Corresponds to: |
Ref document number: 91815 Country of ref document: AT Date of ref document: 19930815 Kind code of ref document: T |
|
REF | Corresponds to: |
Ref document number: 3688737 Country of ref document: DE Date of ref document: 19930826 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: LU Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 19930831 |
|
REG | Reference to a national code |
Ref country code: CH Ref legal event code: PL |
|
ET | Fr: translation filed | ||
NLV1 | Nl: lapsed or annulled due to failure to fulfill the requirements of art. 29p and 29m of the patents act | ||
RAP2 | Party data changed (patent owner data changed or rights of a patent transferred) |
Owner name: SCHLUMBERGER TECHNOLOGIES, INC. |
|
PLBE | No opposition filed within time limit |
Free format text: ORIGINAL CODE: 0009261 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT |
|
26N | No opposition filed | ||
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: FR Payment date: 19980720 Year of fee payment: 13 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: GB Payment date: 19980724 Year of fee payment: 13 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: DE Payment date: 19980729 Year of fee payment: 13 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: GB Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 19990813 |
|
GBPC | Gb: european patent ceased through non-payment of renewal fee |
Effective date: 19990813 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: FR Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20000428 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: DE Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20000601 |
|
REG | Reference to a national code |
Ref country code: FR Ref legal event code: ST |